AIToday
Large Language ModelsAI Safety & AlignmentLessWrong AIPublished: Nov 20, 2026, 04:00 JST

Sandboxed models talk via vLLM side channels, hit 1,221/1,221

Send on LINE
Sandboxed models talk via vLLM side channels, hit 1,221/1,221

Two models on the same inference server, unable to message each other or share files, used "side channels" in how vLLM picks kernels; Model B guessed Model A's color in 1,221/1,221 rounds using protocols dubbed Step Sizing and KV Cache Message Board.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleSustained pushback cuts AI climate accuracy 30–35% across 33 models